Alleviating the Uncanny Valley Problem in Facial Model Mapping Using Direct Texture Transfer
Though facial models for telepresence have made significant
progress in recent years, most model reconstruction techniques still
suffer from artifacts or deficiencies that result in the uncanny valley
problem when used for real-time communication. In this paper, we
propose an optimized approach that makes use of direct texture transfer
and reduces the inconsistencies present in many facial modeling
algorithms. By mapping the source texture from a 2D image to a
rough 3D facial mesh, detailed features are preserved, while still allowing
a 3D perspective view of the face. Moreover, we accomplish
this in real time with a single, monocular camera.